Photonics Leaders Being Sold

Corning and Marvell are presented as pressured AI-photonics beneficiaries despite declines of 40% and 17%, respectively, from their highs. Corning is projected to generate $11.1 billion of sales in 2027 and $14.6 billion in 2028, while Marvell's EPS and revenue are expected to rise 61% and 51% over the next year. Marvell's data-center revenue is forecast to increase from $9.8 billion next year to $23.6 billion by 2029, supporting the article's bullish view that institutional inflows could resume as AI infrastructure spending expands.

Analysis

The investable distinction is not simply “AI optics,” but where each vendor sits in the bandwidth bottleneck. MRVL has higher direct sensitivity to the transition toward 800G/1.6T optical interconnects and custom cloud silicon, while GLW is more exposed to the physical-network build cycle, where customer qualification periods and fiber deployment schedules make revenue recognition slower. This favors MRVL in a near-term hyperscaler capex reacceleration, but GLW if the market rotates from compute spending toward power, connectivity and campus-scale data-center infrastructure over the next 6-18 months.

The important second-order risk is that optics demand can remain strong while vendor economics weaken: module/DSP customers and large cloud buyers can use multi-sourcing to force price concessions once supply catches up. Lumentum (LITE), Coherent (COHR), Fabrinet (FN), and Cisco (CSCO) are relevant read-throughs; broad strength in those names would validate a real optical-content upcycle, while isolated strength in MRVL would more likely reflect company-specific AI expectations. MRVL's earnings sensitivity also carries greater downside because custom silicon programs are concentrated and milestone-driven.

Consensus appears to treat the share-price pullback as a clean entry point, but the missing variable is whether optical attach rates rise fast enough to offset declining unit pricing and any AI-capex digestion. For the next 1-3 months, confirmation should come from hyperscaler capex commentary, MRVL data-center revenue/gross-margin guidance, and LITE/COHR order trends. A downward revision to cloud capex, a sequential deceleration in MRVL's data-center revenue, or GLW backlog conversion slipping would falsify the constructive thesis; neither name should be bought solely on promotional flow indicators.

Key Decisions for Investors

• Prefer a 3-6 month long MRVL / short SOXX pair only after MRVL reaffirms next-quarter data-center growth and gross-margin trajectory. This isolates optical/custom-silicon share gains from a broad semiconductor multiple reset; exit if MRVL guides data-center revenue below consensus or SOXX-relative performance breaks materially after earnings.
• Use GLW as the lower-beta 6-18 month connectivity-infrastructure exposure, but scale in around quarterly results rather than chase a technical bounce. Add only if management demonstrates backlog conversion and pricing stability; the principal risk is delayed data-center construction translating into working-capital build rather than revenue.
• Monitor LITE, COHR, and FN earnings as a confirmation basket before increasing either core position. Positive bookings plus stable gross margins across at least two of these suppliers would support a broader optical-cycle allocation; weak orders despite favorable AI commentary would argue that demand is being double-counted through the supply chain.
• Avoid outright long calls until implied volatility and the timing of hyperscaler capex disclosures are known. If a confirmed catalyst emerges, defined-risk call spreads 3-6 months out on MRVL are preferable to outright calls because a high-growth valuation remains vulnerable to multiple compression even if revenue estimates hold.